It was a method for reducing dimensions by finding the 'direction in which the data spreads the most' through eigenvalue decomposition of the covariance matrix. There is a method similar to PCA called ...